基于振动信号的汽油机爆震边缘检测及诊断研究

Knocking is an abnormal combustion of the gasoline engine. Small amounts of knocking can improve the engine performance, but heavy knocking in engine has some unfavorable effects such as increase in engine pollution and decrease in engine efficiency. Therefore, it is of great significance to find an effective method to extract the knock feature and diagnose the knock intensity exactly. The engine cylinder block vibration signal caused by in-cylinder pressure wave can be measured by accelerometers that mounted outside the cylinder. So this method can be applied for knocking threshold detection. But unfortunately, the components of vibration signal acquired from cylinder block are very complex, which consist of many other vibrational excitation sources and severe noise. Accordingly, traditional fault signal processing and recognition method cannot effectively detect the knock feature and exactly diagnose its intensity. On the basis of the cylinder block vibration signal characteristics, an integrated adaptive signal processing method for knocking threshold detection by synthesizing nonlinear wavelet transform and ensemble empirical mode decomposition is proposed. From the perspective of pattern classification, an attribute reduction model for knocking feature parameters based on neighborhood rough set is constructed to obtain the optimal feature parameter subsets in multi-granular. Combining the diagnosis output results of each sub-classifier in different granularity levels by the means of the criterion matrix algorithm, and then the recognition of engine knocking intensity can be implemented. In a word, we expect to present a set of modeling method to solving these kinds of problem through the research of this project.

爆震是汽油机不正常的燃烧现象之一，轻微爆震可以使发动机性能有所提升，但严重爆震则会影响发动机的运行和排放。因此，准确地进行爆震边缘检测及诊断具有重要的研究意义和工程应用价值。 通过安装在缸体上的振动传感器能够测量出由缸内压力波作用下的机体振动，因此可用于爆震边缘检测。但从缸体上采集的振动信号成分复杂、振源众多且噪声干扰强，从而造成传统的故障信号处理和识别方法无法对爆震特征及其强度进行有效地检测和诊断。 在考虑缸体振动信号特性的基础上，综合非线性小波变换和集合经验模式分解，提出了一种爆震边缘检测的集成自适应信号处理方法；从模式分类角度，建立了采用邻域粗糙集对爆震特征参量进行属性约简的模型，提取出多粒度下分类性能最优的特征参量子集；利用评估矩阵对不同粒度下各子分类器的诊断结果进行融合，实现对爆震强度的分类识别。通过本项目的研究，预期形成一套解决此类问题的模型化方法。

爆震是汽油机不正常的燃烧现象之一，轻微爆震可以使发动机的性能有所提升，但是严重爆震则会影响发动机的运行和排放。因此，准确地进行爆震边缘检测及诊断具有重要的研究意义和工程应用价值。通过安装在缸体上的振动传感器能够测量出由缸内压力波作用下的机体振动，因此可用于爆震边缘检测。但从缸体上采集的振动信号成分复杂、振源众多且噪声干扰强，从而造成传统的故障信号处理和识别方法无法对爆震特征及其强度进行有效地检测和诊断。.本项目以汽油机缸体振动信号为研究对象，以成功检测爆震微弱边缘特征和实现爆震强度智能诊断为研究目标，采用理论研究与试验相结合的方法，对具有信号局部自适应特性的爆震边缘检测方法、基于缸体振动信号的爆震强度指标属性约简模型、微弱爆震故障信号检测和智能诊断识别的模型化方法等研究内容进行了研究和探索。先后研究了非线性冗余第二代小波变换和集合经验模式分解技术在基于振动信号的爆震边缘检测中的应用，成功地从复杂背景信号中检测出微弱的爆震冲击特征波形，验证了两种方法在爆震边缘检测中的适用性和互补性，同时证实了仅利用一只振动加速度传感器来检测多缸发动机多个气缸爆震冲击特征波形的可行性（相关研究成果发表在IOP期刊Measurement Science and Technology上，并被评为“Highlights of 2014”）。在汽油机爆震强度指标属性约简方面，研究了基于邻域粗糙集的特征选择方法，探索了在特征集和决策集之间复杂非线性映射关系下的爆震敏感特征选择效果和分类结果的鲁棒性问题。在汽油机爆震强度智能诊断方面，提出了适用于小样本、不平衡爆震样本集的爆震强度智能诊断的支持向量机方法。上述相关研究成果已经发表SCI/EI 检索的学术论文5篇，在审2篇。.通过本项目的研究，课题组提出了一套完整的基于振动信号的汽油机爆震边缘检测与诊断的模型框架和方法，有效地解决了复杂背景环境下爆震边缘检测和强度诊断的准确性和可靠性问题，扩展机械设备故障诊断的理论与技术，具有较高的学术价值和工程应用前景。